Paper: arXiv 2510.17393

Authors: Kefan Chen, Hussain Ahmad, Diksha Goel, Claudia Szabo

Abstract

Large Language Models (LLMs) have recently gained popularity in stock trading for their ability to process multimodal financial data. However, most existing methods focus on single-stock trading and lack the capacity to reason over multiple candidates for portfolio construction. Moreover, they typically lack the flexibility to revise their strategies in response to market shifts, limiting their adaptability in real-world trading. To address these challenges, we propose 3S-Trader, a training-free framework that incorporates scoring, strategy, and selection modules for stock portfolio construction. The scoring module summarizes each stock’s recent signals into a concise report covering multiple scoring dimensions, enabling efficient comparison across candidates. The strategy module analyzes historical strategies and overall market conditions to iteratively generate an optimized selection strategy. Based on this strategy, the selection module identifies and assembles a portfolio by choosing stocks with higher scores in relevant dimensions. We evaluate our framework across four distinct stock universes, including the Dow Jones Industrial Average (DJIA) constituents and three sector-specific stock sets. Compared with existing multi-LLM frameworks and time-series-based baselines, 3S-Trader achieves the highest accumulated return of 131.83% on DJIA constituents with a Sharpe ratio of 0.31 and Calmar ratio of 11.84, while also delivering consistently strong results across other sectors.

Complexity vs Empirical Score

  • Math Complexity: 2.5/10
  • Empirical Rigor: 6.0/10 (capped from 8.0: transaction costs not modelled; see penalty flags)
  • Quadrant: Street Traders — empirical and applied, lower math complexity (effective empirical rigor capped from 8.0)

Why this score: The paper focuses on an LLM-based framework with minimal advanced mathematical modeling, but presents extensive empirical results with specific performance metrics (returns, Sharpe, Calmar) and backtesting across multiple sectors.

Research Flowchart

  flowchart TD
  A["Research Goal: Adaptive Portfolio Construction using LLMs"] --> B["Data Input: Stock Prices, Financial News, Indicators"]
  B --> C{"3S-Trader Framework"}
  C --> D["Scoring Module: Summarize & Score Stocks"]
  C --> E["Strategy Module: Analyze Market & Optimize"]
  C --> F["Selection Module: Assemble Portfolio"]
  D --> G["Computational Process: Multi-LLM Reasoning"]
  E --> G
  F --> G
  G --> H["Outcome: High Returns & Sharpe Ratio"]
  H --> I["131.83% Return on DJIA (Sharpe 0.31, Calmar 11.84)"]
  H --> J["Consistent Performance Across Sectors"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes 3S‑Trader, a training‑free framework that uses scoring, strategy, and selection modules to construct stock portfolios.

Source sentence
we propose 3S‑Trader, a training‑free framework that incorporates scoring, strategy, and selection modules for stock portfolio construction.

Assumes 1

  • Sharpe ratio is calculated assuming a zero risk‑free rate. statistical
    Source sentence
    assuming a zero risk‑free rate.

Data

Backtest sample: DJIA constituents · equities · weekly

  • Alpha Vantage data for DJIA Constituents · Alpha Vantage market-prices commercial
    May 16, 2022 to May 27, 2024 · DJIA Constituents (30 large‑cap U.S. companies)
    Source sentence
    Data are retrieved from Alpha Vantage 6, and preprocessed according to the methodology detailed in Section 3.2. The evaluation period spans from May 16, 2022 to May 27, 2024.
  • Alpha Vantage data for Technology Sector Stocks · Alpha Vantage market-prices commercial
    May 16, 2022 to May 27, 2024 · Technology Sector Stocks (44 NASDAQ‑100 constituents)
    Source sentence
    Data are retrieved from Alpha Vantage 6, and preprocessed according to the methodology detailed in Section 3.2. The evaluation period spans from May 16, 2022 to May 27, 2024.
  • Alpha Vantage data for Financial Sector Stocks · Alpha Vantage market-prices commercial
    May 16, 2022 to May 27, 2024 · Financial Sector Stocks (49 major companies with highest weights in SPDR Financial Select
    Source sentence
    Data are retrieved from Alpha Vantage 6, and preprocessed according to the methodology detailed in Section 3.2. The evaluation period spans from May 16, 2022 to May 27, 2024.
  • Alpha Vantage data for Healthcare Sector Stocks · Alpha Vantage market-prices commercial
    May 16, 2022 to May 27, 2024 · Healthcare Sector Stocks (46 top‑weighted companies in SPDR Health Care Select Sector)
    Source sentence
    Data are retrieved from Alpha Vantage 6, and preprocessed according to the methodology detailed in Section 3.2. The evaluation period spans from May 16, 2022 to May 27, 2024.

Recognised sources: Alpha Vantage

Costs

Not modeled A trading or execution result is reported without costs.

Finds 2

Reported performance (3S-Trader): Sharpe 0.31 · total return 131.8%

  • 3S‑Trader achieves the highest accumulated return of 131.83% on DJIA constituents, with a Sharpe ratio of 0.31 and a Calmar ratio of 11.84. performance
    Source sentence
    3S‑Trader achieves the highest accumulated return of 131.83% on DJIA constituents with a Sharpe ratio of 0.31 and Calmar ratio of 11.84
  • The method demonstrates stable performance across diverse market environments, consistently ranking among the top two models on all evaluation metrics. performance
    Source sentence
    Our proposed method demonstrates stable performance across diverse market environments, consistently ranking among the top two models across nearly all evaluation metrics.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 17 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (55)

  1. Portfolio Selection (1952)
  2. Textual analysis in accounting and finance: A survey (2016)
  3. Stock market analysis with text data: A review (2021) — link
  4. Simple Overview of Quantitative Analysis (2022) — link
  5. Common risk factors in the returns on stocks and bonds (1993) — link
  6. Giving Content to Investor Sentiment: The Role of Media in the Stock Market (2007) — link
  7. Fundamental Analysis, Future Earnings, and Stock Prices (1997)
  8. Stock market prediction via deep learning techniques: A survey — → 2212.12717
and 47 more
  1. Fingpt: Democratizing internet-scale data for financial large language models (2023) — link
  2. Financial statement analysis with large language models — → 2407.17866
  3. Language models are few-shot learners (2020)
  4. Gpt-4 technical report (2023) — link
  5. On the opportunities and risks of foundation models (2021) — link
  6. Time Series Analysis, Forecasting and Control (1990)
  7. Generalized autoregressive conditional heteroskedasticity (1986)
  8. Deep learning with long short-term memory networks for financial market predictions (2018)
  9. Conditional time series forecasting with convolutional neural networks (2017) — link
  10. Informer: Beyond efficient transformer for long sequence time-series forecasting (2021)
  11. Transformer-based deep learning model for stock price prediction: A case study on Bangladesh stock market (2023)
  12. A multimodal foundation agent for financial trading: Tool-augmented, diversified, and generalist — → 2402.18485
  13. Robust Portfolio Optimization and Management (2007)
  14. Optimizing cyber defense in dynamic active directories through reinforcement learning (2024)
  15. Enhancing network resilience through machine learning-powered graph combinatorial optimization: Applications in cyber defense and information diffusion (2023) — link
  16. Microservice vulnerability analysis: A literature review with empirical insights (2024)
  17. How many stocks make a diversified portfolio? (1987)
  18. Co-Evolutionary Defence of Active Directory Attack Graphs via GNN-Approximated Dynamic Programming (2025) — link
  19. Towards deep learning enabled cybersecurity risk assessment for microservice architectures (2025)
  20. Machine learning driven smishing detection framework for mobile security (2024) — link
  21. Resilient Auto-Scaling of Microservice Architectures with Efficient Resource Management (2025) — link
  22. Towards resource-efficient reactive and proactive auto-scaling for microservice architectures (2025)
  23. Smart HPA: A resource-efficient horizontal pod auto-scaler for microservice architectures (2024)
  24. What Skills Do Cyber Security Professionals Need? (2025) — link
  25. A survey on immersive cyber situational awareness systems (2025)
  26. Chatnvd: Advancing cybersecurity vulnerability assessment with large language models (2024) — link
  27. RegimeFolio: A Regime Aware ML System for Sectoral Portfolio Optimization in Dynamic Markets — → 2510.14986
  28. " I think this is the most disruptive technology": Exploring Sentiments of ChatGPT Early Adopters using Twitter Data (2022) — link
  29. The future of ai: Exploring the potential of large concept models (2025) — link
  30. Active portfolio management: a quantitative approach to providing superior returns and controlling risk (2000)
  31. Deep learning for financial applications: A survey (2020)
  32. Adaptive and explainable margin trading via large language models on portfolio management (2024)
  33. Leveraging large language models and prompt settings for context-aware financial sentiment analysis (2024)
  34. A deep reinforcement learning framework for the financial portfolio management problem (2017) — link
  35. Reinforcement learning application in portfolio optimization: a comprehensive literature review (2024)
  36. Deep direct reinforcement learning for financial signal representation and trading (2016)
  37. Technical Analysis of the Financial Markets: A Comprehensive Guide to Trading Methods and Applications (1999)
  38. Technical Analysis Library in Python (2019)
  39. TradingAgent: LLMs for Trading via Chain of Thought and Memory (2023) — link
  40. Long short-term memory (1997)
  41. Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting (2021)
  42. FinVision: A Multi-Agent Framework for Stock Market Prediction — → 2411.08899
  43. CryptoTrade: A reflective LLM-based agent to guide zero-shot cryptocurrency trading (2024)
  44. Learning to generate explainable stock predictions using self-reflective large language models — → 2402.03659
  45. FinBERT: Financial Sentiment Analysis with Pre-trained Language Models (2019)
  46. Evaluating LLMs’ Mathematical Reasoning in Financial Document Question Answering (2024) — link
  47. FinVision: A multi-agent framework for stock market prediction — → 2411.08899

7 of 55 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.